Precisión diagnóstica de la inteligencia artificial comparada con la interpretación humana en radiografías panorámicas.

Abstract

Background: Artificial intelligence has been incorporated into dental image analysis as a supportive tool for identifying various findings. However, its performance can vary depending on the system used and the type of abnormality being assessed; therefore, comparison with professional interpretation is necessary. Objective: To determine whether the diagnostic accuracy of the Gemini and NeuroVet Dental artificial intelligence systems is comparable or superior to human interpretation in dental panoramic radiographs. Materials and Methods: A quantitative, observational, retrospective study on diagnostic accuracy. A total of 140 digital panoramic radiographs were analyzed using Gemini, NeuroVet Dental, and an rehabilitation specialist. The presence of caries, crown fractures, restorations, implants, and fixed prostheses was evaluated, using the interpretation of an oral radiology specialist as the reference standard. Sensitivity, specificity, accuracy, predictive values, and agreement were calculated. Results: Human evaluation demonstrated the highest accuracy and agreement for the majority of findings. Gemini yielded comparable results for caries, with an accuracy of 94.3% compared to 96.4% for the human evaluator. No significant differences were found between the methods regarding implants (p=0.135). For fractures, restorations, and fixed prostheses, human evaluation was significantly more accurate than both artificial intelligence systems (p<0.001). Conclusion: Gemini and NeuroVet Dental did not achieve diagnostic accuracy comparable or superior to human interpretation. Their performance depended on the specific finding being evaluated; thus, while they can serve as supportive tools, they require professional supervision.